FADU Moves Filesystem into SSD Flash to Bypass CPUs

Summary: FADU’s FS-SSD solution integrates the filesystem directly into the SSD’s flash memory. This allows components like GPUs, DPUs, and NICs to perform file operations without needing the host CPU.

FADU has developed a new FS-SSD (Filesystem-SSD) technology that embeds the filesystem directly into the SSD’s flash memory. This innovation aims to eliminate the overhead traditionally imposed on the host CPU for filesystem operations in conventional storage architectures. I/O bottlenecks are a significant problem, especially in high-performance computing (HPC) and AI/ML workloads.

FADU’s FS-SSD operates by embedding NFSv3 file operations within NVMe commands. This means the SSD’s controller hosts a filesystem layer itself. Consequently, components like GPUs, DPUs (Data Processing Units), and NICs (Network Interface Cards) that interact directly with storage can communicate with the SSD by bypassing the host operating system’s filesystem stack. This reduces latency and increases throughput.

This technology has the potential to significantly boost I/O performance, especially in data-intensive applications. Relieving the host CPU from filesystem overhead allows CPU resources to be dedicated to application workloads. This translates to more efficient resource utilization and faster processing times in AI training, big data analytics, and HPC clusters.

For infrastructure engineers, this offers an opportunity to rethink storage architectures. By reducing the overhead of traditional filesystems, performance gains can be achieved, particularly in next-generation systems requiring direct storage access with DPUs and GPUs. This could be a significant step in the evolution of distributed storage and data fabric solutions.

🔗 Source: Tom’s Hardware
📡 Authoritative feeds & live tracking: Radar